Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 10 May 2002 13:56:45 -0700
From:      Terry Lambert <tlambert2@mindspring.com>
To:        Ramkumar Chinchani <rc27@cse.Buffalo.EDU>
Cc:        freebsd-hackers@freebsd.org
Subject:   Re: kevent subsystem
Message-ID:  <3CDC340D.AC3845DE@mindspring.com>
References:  <Pine.SOL.4.30.0205101424240.2869-100000@pollux.cse.buffalo.edu>

next in thread | previous in thread | raw e-mail | index | archive | help
Ramkumar Chinchani wrote:
> I am asking more in terms of the posix event logging mechanism being
> implemented in Linux 2.5.x kernel.
> 
> http://evlog.sourceforge.net/
> 
> How does the kevent mechanism of event notification and handling compare
> to this scheme?

The POSIX 1003.25 draft that they are referencing:

	http://evlog.sourceforge.net/posix_evlog.html

Is much more comparable to the BEA systems, IBM MQ-Series, or
Web Methods Integration Platform, or, to point to older technology
like I did in my last message, the Tuxedo Transaction Monitor.

The kevent mechanism is an interleaved I/O version of the "select"
or "poll" interface, with bidirectional communications being possible
(e.g. definition of the selection of an accept filter, when registering
for accept events, etc.).  The primary benefit is that, unlike other
mechansims, you can funnel all events through a single multiplexer,
which means that you can handle them without stalls for different
interfaces (though I have yet to see S?YSVIPC intergrated, it's
possible to do).

Event logging is much more of a mechanism for transaction persistance
than it is a replacement for select/poll/kevent.


> It appears to me that the Linux event logging merely supports logging of
> events in a different way that already exists via klogd. This is a more
> passive technique and requiring disk usage overheads in case of huge and
> rapid audits.

I think I would prefer it if you called it "POSIX 1003.25 event
logging", as opposed to "Linux event logging".  Standards are
important, and are cross-platform.

But no.  It's is the basis for a reliable transactioning system.  It
is a fundamental building block for reliable business applications.
Losing an event that encapsulates a $5M transaction is something
which could break your business.


> On the other hand, kevent allows more active event registering and
> handling...
> 
> What would be the pros and cons of both these techniques?

They are totally different problem/solution spaces.

-- Terry

To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-hackers" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?3CDC340D.AC3845DE>